You are on page 1of 7

PHYSICAL REVIEW B 87, 045128 (2013

)
Networks of quantum wire junctions: A system with quantized integer Hall resistance
without vanishing longitudinal resistivity
Jaime Medina,
1,2
Dmitry Green,
3,*
and Claudio Chamon
2
1
Facultad de Ciencias, Universidad Aut ´ onoma de Madrid, 28049 Cantoblanco, Madrid, Spain
2
Physics Department, Boston University, Boston, Massachusetts 02215, USA
3
170 East 83rd Street, New York, New York 10028, USA
(Received 9 October 2012; revised manuscript received 8 January 2013; published 30 January 2013)
We consider a honeycomb network built of quantum wires, with each node of the network having a Y junction
of three wires with a ring through which flux can be inserted. The junctions are the basic circuit elements for
the network, and they are characterized by 3 ×3 conductance tensors. The low energy stable fixed point tensor
conductances result from quantum effects, and are determined by the strength of the interactions in each wire
and the magnetic flux through the ring. We consider the limit where there is decoherence in the wires between
any two nodes, and study the array as a network of classical three-lead circuit elements whose characteristic
conductance tensors are determined by the quantum fixed point. We show that this network has some remarkable
transport properties in a range of interaction parameters: It has a Hall resistance quantized at R
xy
= h/e
2
, although
the longitudinal resistivity is nonvanishing. We show that these results are robust against disorder, in this case
nonhomogeneous interaction parameters g for the different wires in the network.
DOI: 10.1103/PhysRevB.87.045128 PACS number(s): 73.63.Nm
I. INTRODUCTION
The transport properties of junctions of quantum wires are
of interest seen both from basic and applied perspectives.
From the basic physics aspect, quantum wires provide experi-
mentally realizable ways for studying interacting electrons in
one-dimensional geometries, and in particular junctions where
three or more wires meet can display rather rich behaviors.
Theoretically, the problem of quantum wire junctions is
related to dissipative quantum mechanics in two or higher
dimensions, and to boundary conformal field theory.
1,2
It
also has a mathematical connection to certain aspects of
open string theory in a background magnetic field.
3,4
From a
practical viewpoint, junctions of quantum wires should serve
as important building blocks for the integration of quantum
circuits, as they are the natural element to split electric signals
and serve as interconnects.
Junctions of quantum wires have been the subject of many
recent studies
1,2,5–23
which have uncovered many interesting
transport properties as a function of interaction strength.
Quantum wires with few transport channels, at low energies,
can be described as Tomonaga-Luttinger liquids, characterized
by a Luttinger parameter g which encodes the electron-electron
interactions.
24–27
The transport properties of a given junction
depends on the Luttinger parameters for each wire. At low
energies, the conductance properties of the junctions of n wires
are encoded in an n ×n conductance tensor or matrix G
jk
that
relate the incoming currents to the applied voltages on the wires
via I
j
=

k
G
jk
V
k
. At low voltages and low temperatures,
the tensor takes universal forms dictated by the nature of the
infrared stable fixed points in the renormalization group (RG)
sense. These fixed points have been categorized for the case of
Y junctions (n = 3) of spinless
2
and spinful
12
electrons as a
function of the interaction parameter g when all the wires are
identical, and more recently in the case when the wires are not
identical and have different values g
i
.
28
In this paper we investigate the transport properties of
networks constructed using Y junctions of quantum wires as
FIG. 1. (a) Scheme of a grid showing the flow of the current and
the boundary conditions. External currents are fixed, as well as the
potential on the node on the upper right corner. (b) Building block
of the grid: Junction of three quantum wires with a magnetic flux
threading the ring. The V
1,2,3
are the voltages applied on each wire,
and the I
1,2,3
are the currents arriving at the junction from each of the
three wires.
building blocks. Figure 1(a) depicts an example of a network
shaped in the form of a rectangle, and Fig. 1(b) shows the
individual Y junctions used in each node. We consider a
simplified model where the 3 ×3 conductance tensor for each
Y junction is taken to be that dictated by the low energy
quantum RG fixed point, but the transport is treated classically
between any two junctions. The treatment is sensible if the
segment of the wires between two junctions is large compared
to the characteristic dephasing length in the system. But the
length scales of the junction itself, for example, the size of a
ring as shown in Fig. 1(b), should be smaller than the dephasing
length so that the junction is treated quantum mechanically.
The case when the full systemis treated quantummechanically
is extremely difficult to analyze, because it is an interacting
problem. For instance, a lattice version of the problem would
essentially be an example of a two-dimensional interacting
lattice model with a fermion sign problem.
We find rather remarkable results for the transport char-
acteristics of the network of Y junctions, even when the
role of quantum mechanics is just to select the RG stable
fixed point conductances of the elementary building blocks.
045128-1 1098-0121/2013/87(4)/045128(7) ©2013 American Physical Society
JAIME MEDINA, DMITRY GREEN, AND CLAUDIO CHAMON PHYSICAL REVIEW B 87, 045128 (2013)
When the conductance is controlled by the chiral fixed points
χ
±
,
1,2
we find that the whole network behaves as a Hall
bar, with a Hall resistance that is quantized to R
xy
= ±h/e
2
,
like in the integer quantum Hall effect, with the sign given
by the particular chirality of the fixed points χ
+
or χ

.
However, the longitudinal resistivity ρ
xx
= 0, unlike in the
case of the quantized Hall effect where ρ
xx
vanishes. The
quantization of R
xy
is a manifestation of the universal fixed
point conductances. The chiral fixed points are stable for a
range of Luttinger parameters 1 < g < 3, and which of χ
+
or
χ

is selected depends on the flux threading the ring in the
Y junction.
1,2
The flux breaks time-reversal symmetry, but it
does not need to be quantized at any given value; because of
interactions, the conductance of the Y junction flows to fixed
point values for a range of fluxes.
The quantization of R
xy
= ±h/e
2
for the network as a
whole is independent of the value of g in the wires, as long as
they are in the range of stability of the chiral fixed points.
Moreover, we show that the quantization R
xy
= ±h/e
2
is
stable against disorder in the wire parameters. Specifically,
we show that the quantization of R
xy
remains even when the
values of g for different wires are not uniform but disordered,
i.e., they are randomly distributed around some average value
g with some spread δg.
The paper is organized as follows. In Sec. II we briefly
review the results for the conductance characteristics of single
quantumYjunctions, which are the elementary building blocks
for the honeycomb wire networks. In Sec. III we present
analytical results from which one can understand the origin of
the quantization of R
xy
when the conductance tensor of each
of the Y junctions in the network is associated to a chiral fixed
point. In Sec. IV we present numerical studies confirming the
analytical findings by analyzing grids with different values of
the interaction parameter g, different geometries and sizes, and
extrapolate these results to the thermodynamic limit. These
numerical calculations are of much value for the next step,
taken in Sec. V, where we discuss the robustness of the
quantization of the Hall resistance in the case when the wires
each have different Luttinger parameters distributed randomly.
The Appendix contains a detailed description of the numerical
method to solve our network of Y junctions.
II. SINGLE Y JUNCTION AS AN ELEMENTARY
CIRCUIT ELEMENT
Each of these Y junctions in the network consists of three
wires that are connected to a ring which can be threaded by
a magnetic flux, as shown in Fig. 1(b). This flux breaks time-
reversal symmetry, and the currents in the junction will depend
on the potential at its extremes and the magnetic flux inside
the junction.
The current-voltage response of each Y junction is deter-
mined by its conductance tensor G
jk
. Within linear response
theory, the total current I
j
flowing into the junction from wire
j is related to the voltage V
k
applied to wire k by
I
j
=

k
G
jk
V
k
, (2.1)
where j,k = 1,2,3. Two sum rules apply to the conductance
tensor because of conservation of current and because the
currents are unchanged if the voltages are all shifted by a
constant:

j
G
jk
=

k
G
jk
= 0. (2.2)
The G
jk
reach universal values at lowtemperatures and low
bias voltages. These universal values are dictated by the RG
stable fixed point that is reached for given values of the
Luttinger parameters in the wires. Here we shall focus on
the case where all three wires have the same parameter g. In
Sec. V we will consider the more general case of a network of
wires where the three wires for each Y junction have different
g.
When the three wires have the same g, the fixed point
conductance tensor has a Z
3
symmetry and takes the form
2
G
jk
=
G
S
2
(3δ
jk
−1) +
G
A
2

jk
, (2.3)
where
ij
= δ
i,j−1
−δ
i,j+1
with i +3 ≡ i and we separate
the symmetric and antisymmetric components of the tensor,
whose magnitudes are encoded in the scalar conductances G
S
and G
A
. G
A
vanishes when time-reversal symmetry is not
broken, for instance, in the absence of magnetic flux through
the ring.
The fixed point values of G
S
and G
A
depend on the strength
of electron-electron interactions, encoded in the Luttinger
parameter g. We will focus on the chiral fixed points χ
±
,
which are stable in the range 1 < g < 3.
1,2
In the chiral
cases, the conductances are given by G
S
= G
χ
=
e
2
h
4g
3+g
2
and
G
A
= ±g G
χ
. Thus the chiral conductance tensors are
G
±
jk
=
G
χ
2
[(3δ
jk
−1) ±g
jk
]. (2.4)
We shall work in units where the quantum of conductance
e
2
/h is set to 1.
The Y junctions are then assembled into a network as
shown in Fig. 1(a). We consider a regular hexagonal grid of
Y junctions with 2c external connections on both the top and
bottomsides and r on both the right and left side. Parametrized
in such a way and with wires of unit length, the dimensions of
the grid as a function of r and c are
L
x
= 6c,
(2.5)
L
y
=

3(2r +1).
In this grid we shall fix the current flow along the x axis
from left to right and we shall fix the currents flowing into
the top and the bottom to zero, as shown in Fig. 1(a). Given
the conductance tensors at every node of the network, we
compute the potentials and the currents on the links of the grid.
The resistances and resistivities of the networks are studied
for different orientations and systems sizes, and for different
values of g. In the Appendix we present details of the method
used to numerically compute the response of the networks.
III. ANALYTICAL RESULTS
We will measure the longitudinal and transverse responses
in the framework of the classical Hall problem by injecting a
045128-2
NETWORKS OF QUANTUM WIRE JUNCTIONS: A SYSTEM . . . PHYSICAL REVIEW B 87, 045128 (2013)
transverse current along the x axis and imposing a zero current
boundary condition along the two edges that are parallel to the
x axis. This approach suggests that we solve for the potential
in the bulk as a function of the external current. In other words,
we need to invert the fundamental equation (2.1) for I and V
for each junction in the bulk.
While the full network problemis not tractable analytically,
we can still gain some insight from a combination of
analytics and heuristics. In particular, we will be able to prove
quantization of the transverse resistivity analytically, even with
some forms of disorder. Similarly we will derive the general
form of the longitudinal resistivity. We will confirm these
results numerically in later sections. Let us start with the unit
cell of the hexagonal lattice. There are two vertices (nodes)
in each cell and current is directed along the bonds (wires)
as shown in Fig. 2. Looking at the right-hand node first, the
potentials on the external wires V
2
and V
3
, and the potential
on the internal wire V
1
, are defined only up to an additive
constant. This means that Eq. (2.1) is not invertible. However,
by setting V
1
= 0, or equivalently shifting all potentials in the
two nodes by a constant V
i
→V
i
−V
1
, the gauge is fixed and
we obtain, using Eq. (2.4), the following:

V
2
−V
1
V
3
−V
1

=
1
2g

2 1 ∓g
1 ±g 2

I
2
I
3

. (3.1)
The solution in the left node is similar but with the permutation
(V
2
,V
3
) →(V

2
,V

3
) and (I
2
,I
3
) →−(I

2
,I

3
), which follows
from rotational symmetry and the orientation that we have
chosen for the currents.
Now consider the potential gradient in the x and y
directions. It is straightforward to derive the change in potential
per unit cell V
x
and V
y
directly from Eq. (3.1) as follows:

x
V = V

3
−V
2
=
1
2g
[2I
1
−I
2
−I

3
∓g(I

2
−I
3
)],
(3.2)

y
V = V
2
−V
3
=
1
2g
[±gI
1
+I
2
−I
3
].
It is instructive to consider a simple case. We will generalize
this result below, but for now consider a uniform current in
the bulk in the x direction (or “armchair” configuration to
borrow nomenclature from graphene). Each horizontal wire in
V
V
V
3 2
2 3
V
A D
C
B
I
I
I
I
I
2
3
1
3
2
FIG. 2. Unit cell of the hexagonal network. Currents are assumed
to be positive when directed along the arrows in the wires. Dotted
lines denote the boundary of the unit cell. The rectangular region
ABCD shown is used for computing the resistances and resistivities
of the network.
each unit cell has a current I
1
= I. By symmetry the other
wires split the current equally: I
2
= I
3
= I

2
= I

3
= −I/2.
This configuration leads to a particularly simple potential
gradient:
x
V = 3I/2g and
y
V = ±I/2.
The result for the resistances and resistivities are apparent
after we account for the geometric factors. Consider the
rectangular region ABCD in Fig. 2, with sides d
AB
=

3 and
d
AD
= 2. In the transverse direction the width of the rectangle
is twice the distance between the midpoint of the wires (with
currents I
2
and I
3
), and the voltage drop V
AB
= 2
y
V. The
Hall resistance (which coincides with the Hall resistivity ρ
xy
)
is therefore R
xy
= V
AB
/I = 2
y
V/I = ±1. In other words
the Hall resistance is independent of g and quantized to unity!
Similarly, in the longitudinal direction the length of the
rectangle is 4/3 the distance between the midpoint of the wires-
(with currents I

2
and I
3
), and V
AD
= 4/3
x
V. The longi-
tudinal resistance is R
xx
= 4/3
x
V/I = 2/g. There is an
additional geometric factor in the longitudinal resistivity given
by ρ
xx
= (d
AB
/d
AD
) R
xx
, and it is thus given by ρ
xx
=

3/g.
Hence the resistivity is nonzero and there is dissipation unlike
in the standard quantum Hall effect.
Had we used an alternate (“zigzag”) configuration where
the transverse current is zero I
1
= 0 and the uniform current is
in the y direction, I
3
= I

3
= −I
2
= −I

2
= I, we would have
found a similar result, i.e., that the resistance in the x direction
is quantized to R
xy
= ±1 while the resistivity in the y direction
is ρ
yy
=

3/g.
We find this result both unexpected and remarkable. By
taking the classical conductivity limit for each wire we
have allowed decoherence along the wires. However we
have preserved the quantum coherence on each vertex, as
the chiral relation Eq. (2.4) is by nature a consequence
of quantum scattering. Nonetheless, even after relaxing a
portion of the coherence, some element of quantization in the
thermodynamic limit has survived in the form of an integer
quantized Hall resistivity. On the other hand, decoherence has
destroyed the zero longitudinal resistivity of the quantum Hall
effect, and so we are left with a hybrid quantum-classical
Hall effect. Note also that the simple uniform solution above
suggests robustness against disorder, another element of the
integer quantum Hall effect. As the transverse gradient of V
is independent of g in the uniform bulk, suppose that g is
allowed to vary slowly from vertex to vertex, more slowly
than the current. In this regime we would expect quantization
to persist, and indeed we will confirm that numerically later in
this paper.
We will substantiate the assumptions and findings above
numerically in the next section.
IV. NUMERICAL RESULTS
In this section we shall present numerical results for the
voltages and currents in the wires of the network. These
numerical studies serve first as a check of the analytical results
presented in Sec. III for the case where all the interaction
parameters are the same for all wires. Second, and more
importantly, they serve as a stepping stone to the case of
nonhomogeneous (disordered) interaction parameters in the
wires, which will be considered in Sec. V. The method used
045128-3
JAIME MEDINA, DMITRY GREEN, AND CLAUDIO CHAMON PHYSICAL REVIEW B 87, 045128 (2013)
to solve for the voltages and currents in the grid is presented
in the Appendix.
Let us focus on the armchair layout of Fig. 1(a) (similar
results follow in the case of the zigzag case). Also, without
loss of generality, we consider below only the χ
+
fixed point.
Current is injected and collected uniformly into the wires on
the right and on the left of the network, respectively. More
precisely, there are r wires serving as connections to the outside
on each side of the grid, and current I = I
x
/r is injected in and
collected out of these external wires. The total current flowing
along the horizontal or x direction is therefore I
x
.
The distribution of the currents in the inner parts of the grid
that follow from this uniform injection of external currents is
shown in Fig. 3. We find a close to uniform distribution, with
slightly larger currents closer to the edges. This distribution is
independent of the value of g. These patterns of current flow
in the inner wires of the grid are in agreement with the current
distributions discussed in the analytical studies of the previous
section.
The Hall voltage is the potential dropV
y
along the vertical or
y direction. We note that the potential drop V
y
is computed by
looking at the potentials for two points at the same horizontal
position (i.e., the same x position), one at the top and one at
the bottom of the network.
We show in Fig. 4 the potentials measured at the top and
at bottom of the (rectangular shaped) grid. Notice that the
potentials drop linearly with the horizontal direction, but that
the difference between the two potentials V
y
is constant.
The Hall resistance is computed as follows. Let
¯
V
y
be
the average over the horizontal positions x of the Hall
voltage drop. (Since in this case without disorder V
y
is
constant, the average is actually unnecessary here.) Then the
Hall resistance is given by R
xy
=
¯
V
y
/I
x
. We find numerically
that R
xy
= 1 as expected from the analytical arguments.
29
Recall that we are working in units where e
2
/h = 1, so indeed
FIG. 3. (Color online) Currents flowing through the Y junctions
that lie along a vertical line in the middle of the bar (x = L
x
/2) as a
function of vertical position y/L
y
. Note that for y values away from
the edges the currents tend to I
1
= 1 and I
2,3
= 1/2, as predicted
analytically for the asymptotic limit.
FIG. 4. (Color online) Voltages at the top and and bottomedges as
a function of horizontal position x/L
x
when the node at the top right
corner is grounded. The grid size is r = 50, c = 60, and g =

3.
Notice that the difference between the voltages at the top and bottom
edges for a given x/L
x
is exactly 1 in natural units.
we have
R
xy
=
h
e
2
, (4.1)
which we find is independent of the value of g. We remark that
we find that this quantization holds independent of the aspect
ratio, orientation (armchair vs zigzag), or size of the grid.
We also computed the potential difference between points
on the left and on the right sides of the grid V
x
as a function of
the vertical direction y. In this case we find that the horizontal
potential difference is almost constant as a function of y (as
opposed to the case of the vertical drop V
y
, which is exactly
independent of x). The difference is bigger, by an amount of
order 1/L
y
, when y is in the middle of the grid as compared to
when y is at the edges. We define
¯
V
x
as the y-position averaged
voltage difference between the left and right sides of the grid.
The longitudinal resistance is given by R
xx
=
¯
V
x
/I
x
, and the
longitudinal resistivity by ρ
xx
= L
y
/L
x
¯
V
x
/I
x
.
We find that the longitudinal resistance is nonzero, in
agreement with Sec. III. We find numerically, however, that
there are finite system size corrections to the analytical
predictions. We find that
R
xx
(g,L
x
,L
y
) =

3
g
L
x
L
y
−A(L
x
,L
y
)
, (4.2)
where A is a factor of order 1 that corrects for finite sizes.
We find numerically that in the thermodynamic limit A →1
for the armchair configuration, whereas A = 0 independent of
system size in the zigzag case. Therefore, in the thermody-
namic limit we obtain
ρ
xx
= lim
L
x
,L
y
→∞
L
y
L
x
R
xx
(g,L
x
,L
y
) =

3
g
, (4.3)
in agreement with the result in Sec. III.
The Hall angle θ
H
is given by tan θ
H
= ρ
xy

xx
, and we
naturally find, given the agreement with the results for ρ
xx
and
045128-4
NETWORKS OF QUANTUM WIRE JUNCTIONS: A SYSTEM . . . PHYSICAL REVIEW B 87, 045128 (2013)
FIG. 5. (Color online) Density plot for the voltages on the grid
nodes for a system with r = 50, c = 60, and g =

3. Notice the
constant slope of the equipotential lines, which is related to the Hall
angle θ
H
. The Hall angle depends on the interaction strength and is
given by Eq. (4.4).
ρ
xy
above, that
tan θ
H
=
g

3
(4.4)
in the thermodynamic limit. This Hall angle can be visualized
very naturally by plotting the voltages at the wires on the grid,
as shown in Fig. 5. The Hall angle appears as the slope of the
lines of constant voltage. These equipotential lines are straight
in this example where all the wires have the same interaction
parameter g; this is no longer the case when disorder is
introduced in Sec. V.
V. ROBUSTNESS AGAINST DISORDER
In this section we will generalize the wire networks to
the case when the interaction parameters g for each of
the wires in the network are not uniform, but instead are
drawn independently from a distribution. We shall consider
a distribution in which g in each of the wires in the network
takes a value between ( ¯ g −δg, ¯ g +δg), with uniform proba-
bility. Because the interaction parameter should be positive,
δg < ¯ g.
When the wires connecting to a given Y junction have
different values of g, the conductance tensor G
jk
for a chiral
fixed point is no longer given by Eq. (2.4), but instead it takes
the form (see Ref. 28)
G
jk
= 2
g
j
(g
1
+g
2
+g
3

jk
+g
j
g
k
(±g
m

jkm
−1)
g
1
g
2
g
3
+g
1
+g
2
+g
3
. (5.1)
Using this conductance tensor, one can compute numerically
(using the method of the Appendix) the voltages and currents
in all wires of the network for a given realization of the
disorder.
We shall show below that the quantization R
xy
= 1 of
the Hall conductance that we found in the clean limit
remains in the thermodynamic limit, even in the presence
of disorder. For a finite lattice, as one should expect, there
are fluctuations that we quantify below for the armchair
configuration.
FIG. 6. (Color online) Standard deviation of the Hall resistance
for 100 simulations with ¯ g =

3 and δg = ¯ g/10 as a function of 1/L
for a grid with r = c = L (which fixes the aspect ratio). It scales to
zero in the large L limit, implying that the system is self-averaging
and the Hall resistance R
xy
→1 independent of disorder in the
thermodynamic limit.
We compute Hall resistance R
xy
(defined as the average of
the voltage differences between top and bottomof the network,
divided by the injected current) for several realizations of
disorder and system sizes. For a fixed system size we
then find the disorder average R
xy
and standard deviation
R
xy
=

R
2
xy
−R
xy
2
of R
xy
. We find that R
xy
→1 as the
number of realizations increase, and that the standard deviation
R
xy
→0 as L increases (we use lattices with r = c = L).
We show in Fig. 6 the finite size scaling of the R
xy
.
That R
xy
→0 in the thermodynamic limit means that the
system is self-averaging, and therefore R
xy
→1 independent
of disorder in the thermodynamic limit. We conclude then that
quantization is robust against disorder.
We have also checked the effects of disorder for the zigzag
configuration, reaching similar conclusions that disorder does
not alter the quantization of the conductance in the thermody-
namic limit.
In summary, we find that, in the thermodynamic limit,
the general results of the previous sections hold even in the
presence of disorder.
VI. CONCLUSIONS
We investigated the transport properties of hexagonal
networks whose nodes are Y junctions of quantum wires.
In our model the 3 ×3 conductance tensor for each Y
junction is dictated by the low energy RG fixed point, but
the transport is treated classically between any two junctions.
We find a surprising result: In spite of relaxing quantum
coherence between the junctions, we find a quantized Hall
resistance.
Specifically, in the regime where the junction conductance
is controlled by the chiral fixed points χ
±
1,2
(when the
045128-5
JAIME MEDINA, DMITRY GREEN, AND CLAUDIO CHAMON PHYSICAL REVIEW B 87, 045128 (2013)
interaction parameter obeys 1 < g < 3), the network exhibits
a quantized Hall resistance R
xy
= ±h/e
2
. This quantization is
similar to that in the integer quantum Hall effect. Furthermore,
the quantization is independent of the interaction parameter
g even in the presence of disorder in g. The quantization
of the Hall resistance follows from the specific form of the
conductance tensor at the RG stable chiral fixed point at each
Y junction. However, unlike in the quantized Hall effect,
where the longitudinal resistivity vanishes, ρ
xx
is not zero:
ρ
xx
= (

3/g) h/e
2
. Dissipation in the longitudinal direction
is a result of decoherence within the wires. We emphasize
that in our model the wires are classical, but the nodes remain
quantum mechanical and the form of the conductance tensor
G at each junction is constrained by quantum scattering
effects. The essential ingredient for the quantization of the Hall
conductance is the value of the chiral fixed point conductance
of the individual junctions.
Finally, let us comment on the finite temperature corrections
to the value R
xy
= ±h/e
2
in the network. As opposed to
the case of the quantum Hall effect where the quantization
is exponentially accurate because of an energy gap, the
quantization in the networks has a power law correction in
T because the wire networks are gapless. The quantization
should be as accurate as the conductance tensor is close to
that of the RG fixed point. The corrections to the conductance
tensor scale as T

, where = 4g/(3 +g
2
) is the scaling
dimension of the leading irrelevant operator at the chiral fixed
points.
1,2
Notice that the temperature scaling of the conductivity
above should hold only under the assumption of decoherence
within the wires. However, as temperature goes to zero, the
coherence length increases, and therefore there is an implicit
assumption of order of limits for the results in this paper to
work as presented: The length of the wires should be taken to
infinity before the limit of T = 0 is taken. But it is natural
to wonder whether the quantization that we found in this
work should persist or not even if transport along the wires is
always coherent. Indeed, one possibility is that in the coherent
regime one might have quantization of the Hall conductance
with vanishing longitudinal resistivity. However, to address
this regime one would need to tackle the fully interacting
two-dimensional fermionic model, which is beyond the scope
of this paper. One route to follow could be to consider a
lattice model where the wires are described by a tight-binding
model, with three wires coupled together at junctions by
hopping matrix elements between them. One could possibly
start with a noninteracting version of the model, where the
chiral conductances used in this paper are obtained by fine
tuning to the fixed point (since the noninteracting model is
marginal and there is no RG flow). The problem then becomes
one of electrons in a superlattice, with the number of bands
scaling with the number of sites describing the wires within a
supercell. The Hall conductance for this tight-binding model
could be obtained by computing the Chern number of the filled
bands. If the Hall conductance does not vanish in this model, it
is only protected algebraically in temperature, as there would
be “mini gaps” separating bands that scale inversely with the
size of the wires, instead of true band gaps. Analyzing such
model may shine some light on the problem of wire networks
in the coherent regimes.
FIG. 7. Example with r = 2, c = 2. Note the row of “ghost
nodes” at the top edge.
ACKNOWLEDGMENT
This work was supported in part by the DOE Grant No.
DE-FG02- 06ER46316 (C.C.).
APPENDIX: METHOD
Our numerical approach consists of solving the full lattice
model exactly. In this Appendix we describe our methodology
in detail.
Consider an arbitrary lattice with r external wires on each
side and 2c external wires at the top and bottom edges. An
equal current will be injected into each of the r wires on the
left, and the 2c edge wires will have a current of zero. This
defines the boundary conditions. A 2 ×2 lattice is shown for
example in Fig. 7. For later convenience we include a row
of 2c “ghost nodes,” shown as dotted lines at the top edge,
but they are only there to facilitate the numbering scheme and
no current will flow through them. Including the ghost nodes
there are a total of N = 4c(r +1) nodes.
The points on each wire that emanate from each node are
governed by the equation V = GI, where Gis a 3 ×3 matrix.
Thus we start with 3N degrees of freedom. However, starting
in this way introduces many redundant variables in the bulk
because in a classical wire the current is the same everywhere
along the wire and so is the potential. We will unify the two
points on each wire in the bulk by imposing a set of constraints.
In general there are 6cr +c −r such constraints, which equals
the number of wires in the bulk.
To write down the full network equation let us label each
of the 3N points by (n,i), where n = 1, . . . ,N is the node
index and i = 1,2,3 refers to the point on each wire that
emanates from each node. The potentials and currents at each
of these points are denoted by V
(n)
i
and I
(n)
i
, respectively. To
illustrate this notation, in Fig. 7 the constraint along the wire
that connects nodes 6 and 7 would be written as V
(6)
1
= V
(7)
1
and I
(6)
1
= −I
(7)
1
.
045128-6
NETWORKS OF QUANTUM WIRE JUNCTIONS: A SYSTEM . . . PHYSICAL REVIEW B 87, 045128 (2013)
Each node obeys the relation V
(n)
= G
(n)
I
(n)
, where G
(n)
is
the 3 ×3 matrix from Eq. (2.1). The network is thus described
by the following linear equation with constraints:





I
(1)
I
(2)
.
.
.
I
(N)





=





G
(1)
0 · · · 0
0 G
(2)
· · · 0
.
.
.
.
.
.
.
.
.
.
.
.
0 0 · · · G
(N)










V
(1)
V
(2)
.
.
.
V
(N)





. (A1)
Next we impose the constraints to reduce the effective
dimensionality of the problem. Start with the set of point
pairs on each wire in the bulk {(n,i),(m,i)}, where n and m
are nearest neighbor nodes. The constraints are V
(n)
i
= V
(m)
i
and I
(n)
i
= −I
(m)
i
for each pair. We impose the constraint on
voltages by adding the (n,i)th and (m,i)th columns together,
removing the (m,i)th column and removing V
(m)
i
from the
vector of potentials in Eq. (A1). Similarly we impose the
constraint on currents by adding the (n,i)th and (m,i)th rows,
deleting the (m,i)th row and removing I
(m)
i
from the vector
of currents. Also we replace the current I
(n)
i
that has not been
eliminated by zero because I
(n)
i
+I
(m)
i
= 0. Therefore each
constraint is equivalent to removing one row and one column
and reduces the dimensionality of the original problemby one.
Furthermore, we have replaced each current in the bulk by zero
which is important because the only currents that are left in
Eq. (A1) are fully determined, being equal to either zero in the
bulk or to the boundary conditions.
Eliminating the ghost nodes is straightforward—we simply
remove the ghost currents, potentials, and their associated rows
and columns in Eq. (A1). This reduces the dimensionality
further by 3 ×2c, which is the number of wires emanating
from the ghost nodes. The final step is to fix the gauge. Since
all potentials are determined up to an overall constant, we pick
an arbitrary potential, set it to zero, and remove the associated
row and column from Eq. (A1).
To summarize, we started with 3N = 12c(r +1) redundant
degrees of freedom and then through successive transfor-
mations we imposed 6cr +c −r constraints in the bulk,
eliminated 6c ghost points, and fixed one potential to zero. The
dimensionality has thus been reduced to 6rc +5c +r −1 and,
crucially, the only currents appearing are either zero or fixed
by boundary conditions. Having eliminated all redundancies
allows us to solve for the potential at any point, as a function
of the boundary currents, by inverting the reduced version of
Eq. (A1), which we do numerically.
The generalization to random couplings g is straightfor-
ward. The derivation proceeds in exactly the same way as we
just described, but we start with nonuniform G
(n)
.
*
dmitrygreen2009@gmail.com
1
C. Chamon, M. Oshikawa, and I. Affleck, Phys. Rev. Lett. 91,
206403 (2003).
2
M. Oshikawa, C. Chamon, and I. Affleck, J. Stat. Mech.: Theory
Exp. (2006) P02008.
3
C. G. Callan, I. R. Klebanov, J. M. Maldacena, and A. Yegulalp,
Nucl. Phys. B 443, 444 (1995).
4
C. G. Callan and D. Freed, Nucl. Phys. B 374, 543 (1992).
5
C. Nayak, M. P. A. Fisher, A. W. W. Ludwig, and H. H. Lin, Phys.
Rev. B 59, 15694 (1999).
6
S. Lal, S. Rao, and D. Sen, Phys. Rev. B 66, 165327 (2002).
7
S. Chen, B. Trauzettel, and R. Egger, Phys. Rev. Lett. 89, 226404
(2002).
8
K.-V. Pham, F. Pi´ echon, K.-I. Imura, and P. Lederer, Phys. Rev. B
68, 205110 (2003).
9
S. Rao and D. Sen, Phys. Rev. B 70, 195115 (2004).
10
K. Kazymyrenko and B. Douc¸ot, Phys. Rev. B 71, 075110 (2005).
11
B. Bellazzini, M. Mintchev, and P. Sorba, J. Phys. A 40, 2485
(2007).
12
C.-Y. Hou and C. Chamon, Phys. Rev. B 77, 155422
(2008).
13
S. Das and S. Rao, Phys. Rev. B 78, 205421 (2008).
14
A. Agarwal, S. Das, S. Rao, and D. Sen, Phys. Rev. Lett. 103,
026401 (2009).
15
B. Bellazzini, P. Calabrese, and M. Mintchev, Phys. Rev. B 79,
085122 (2009).
16
B. Bellazzini, M. Mintchev, and P. Sorba, Phys. Rev. B 80, 245441
(2009).
17
I. Safi, arXiv:0906.2363.
18
D. N. Aristov, A. P. Dmitriev, I. V. Gornyi, V. Y. Kachorovskii, D. G.
Polyakov, and P. W¨ olfle, Phys. Rev. Lett. 105, 266404 (2010).
19
M. Mintchev, J. Phys. A 44, 415201 (2011).
20
D. N. Aristov, Phys. Rev. B 83, 115446 (2011).
21
D. N. Aristov and P. W¨ olfle, Phys. Rev. B 84, 155426 (2011).
22
C. Wang and D. E. Feldman, Phys. Rev. B 83, 045302 (2011).
23
V. Caudrelier, M. Mintchev, and E. Ragoucy, arXiv:1202.4270.
24
S. Tomonaga, Prog. Theor. Phys. 5, 544 (1950).
25
J. M. Luttinger, J. Math. Phys. 4, 1154 (1963).
26
D. C. Mattis and E. H. Lieb, J. Math. Phys. 6, 304 (1965).
27
F. D. M. Haldane, J. Phys. C 14, 2585 (1981).
28
C.-Y. Hou, A. Rahmani, A. E. Feiguin, and C. Chamon, Phys. Rev.
B 86, 075451 (2012).
29
The value R
xy
= 1 that is found numerically is exact to double
precision in Matlab.
045128-7